Six words, delivered at his first meeting, that the market read as the end of the easing bias. In the same room the Fed cut its statement to 130 words, dropped forward guidance, and let the projections carry the hawkish message. Across the Atlantic, Christine Lagarde told Sintra that monetary policy had gone "back to basics"; the ECB's June hike, she said, was "justified under every scenario."

The disinflation era had a reflex. When growth wobbled or a shock hit, central banks cut, and markets learned to buy the dip because the cut was coming. In June 2026 that reflex was withdrawn, and it was withdrawn in concert.

On 11 June the European Central Bank raised its deposit rate to 2.25%, its first hike since 2023, blaming a Middle East war that had "generated inflation pressures." On 16 June the Bank of Japan lifted to 1.00%, a level not seen since 1995. On 17 June the Federal Reserve held at 3.50-3.75%, but its new Summary of Economic Projections lifted the 2026 median dot to 3.8%, above the current band, and nine of eighteen participants pencilled in a hike. The same day the Bank of England held at 3.75% on a 7-2 vote, with two members already wanting to raise.

Only the Bank of Canada, holding at 2.25% for a fifth straight meeting, and the Reserve Bank of India, holding at 5.25%, stayed out of the move. The centre of gravity had shifted decisively hawkish, and it did so in a week rather than over the many months such a turn normally takes.

What made this a regime change rather than a coincidence was the SEP itself. In March the Fed's median dot implied a 2026 cut. By June it implied a hike, a 40-basis-point swing in the central projection, driven by a PCE inflation forecast revised up almost a full point to 3.6%. The market had spent the first half of the year pricing the old reflex. In one afternoon that repricing reversed.

The equity puzzle

And yet developed-market equities did not break. The S&P 500 fell only 1.2% on the month and is up 9.6% on the year; Europe and Japan printed records: the Euro Stoxx 50 at an all-time close, the Nikkei up 5.3% on the month and roughly 37% on the quarter, its best since 1965. Q2 was the best quarter for US stocks since 2020. Risk assets, in other words, shrugged at the very regime the bond market and the central banks were confirming.

Two readings compete. The benign one: growth is genuinely solid (US GDP +2.1%, jobs beating), earnings are carrying the load, and equities can live with higher-for-longer if the "for-longer" reflects strength. The uneasy one: this is the late-cycle pattern in which the last thing to reprice is the riskiest thing, and June's cracks (a 10% one-day chip rout, the Mag 7 turning negative on the year) are the first tremors.

The choreography matters. The ECB and BoJ moved before the Fed, so by the time Chair Warsh stepped to the podium the hawkish direction was already set, and the Fed confirmed it rather than setting it. Then, hours after the decisions, the US-Iran ceasefire reopened the Strait of Hormuz and sent oil into its steepest monthly fall since March 2020. The classic trigger for a dovish reversal arrived immediately, and not one of the four banks blinked. By Sintra on 29 June, Lagarde was describing the hike as "robust across every scenario," and the BIS annual report was praising "policy discipline."

The two that hiked

The ECB is the purest expression of the turn. Euro-area headline inflation had climbed to 3.2%, its highest since 2023, with energy up nearly 11%; core rose to 2.6%. The Governing Council raised all three rates and revised its 2026 inflation path up on "a higher path for energy prices." Lagarde's Sintra coda, that policy is "back to basics" and forward guidance "not in the cards," was the plainest statement of intent all month.

The BoJ hiked to 1.00% on a 7-1 vote, its highest since 1995, and signalled more to come as the yen sat near four-decade lows. Japan's own inflation is modest (headline 1.5%), but a collapsing currency imports it, and the Bank has decided that defending the yen now matters as much as supporting demand.

The holds that were hawkish

The Fed held but did the work through projections; the BoE held 7-2 with Greene and Pill dissenting for a hike; the BoC held for a fifth time but flagged oil "roughly $10 above" its April assumptions. Even the pauses leaned tight.

The oil chart of 2026 is a mountain. War from late February closed the Strait of Hormuz (roughly a fifth of the world's seaborne oil and LNG) and drove Brent above $120 by April. A 17 June US-Iran ceasefire and a toll-free reopening reversed it; Brent ended June near $73, down about 21% on the month, its worst since March 2020, with WTI around $70.

The instinct is to read the round-trip as "shock over, cut coming." Central banks read it differently, for three reasons. First, the spike still fed a real inflation impulse now embedded in the data: US CPI at a three-year 4.2%, euro-area energy up 11%, UK services inflation re-accelerating. Second, the reopening is fragile: renewed strikes on 29 June threatened it, and European gas storage sits near 48%, the lowest heading into winter in roughly 15 years. Third, and most important, this was the second energy shock in three years. When the same chokepoint keeps generating inflation, the credibility of "transitory" is gone.

June was less a risk-off month than a rotation month. Europe and Japan made records: the Nikkei's ~37% quarter was its best since 1965, Euro Stoxx 50 closed at an all-time high on the AI-and-rearmament trade, MSCI EM logged its best quarter since 2009 on a weaker-dollar first half. The pain concentrated in three places: Chinese equities (Hang Seng −9.1%, its worst month of the year, on an AI selloff), US mega-cap tech (the Mag 7 shed ~$2tn), and the disinflation hedges (gold −11.6%, silver −22%, bitcoin −20.3%).

A rate hike is supposed to lift a currency. In June it did not, because everyone hiked. The Bank of Japan raised to a 31-year high and the yen still slid to 162 per dollar, a four-decade low, because a 25-basis-point move buys nothing when the whole field is moving. Tokyo, having already spent an estimated ¥11.7tn (~$73bn) on intervention in April-May, signalled it may return.

The dollar, by contrast, rose 2% to 101.3, its best month since mid-2025, breaking above 100 after the FOMC. Treasury Secretary Bessent used an Economic Club of New York address to underline the point: "There is nothing accidental about the dollar's place in the world." In a higher-and-energy-scarce regime, the reserve currency with the deepest markets and the highest real rate is the default winner.

Emerging-market currencies bore the mirror image. The lira topped 46 for the first time; the real sat near 5.18 even after a Brazilian hike to 14.25%. And the rupee, our marquee case, closed near 94.6, off its 96.9 May record, steadied by an oil that finally fell and by a central bank that chose not to chase the global tightening.

The regime break has a mirror. As the Fed, ECB, BoJ and BoE tightened, the People's Bank of China cut: it held its 1-year loan prime rate at a record-low 3.00%, trimmed the reverse-repo to 1.40%, and lowered reserve requirements. It is the cleanest policy divergence on the board: the West is confronting inflation that will not fade, while China still confronts a shortfall of demand.

The nuance matters, because China's headline prices have turned positive: CPI +1.2%, PPI +3.9% in May, the first factory-gate inflation since 2022. But that is the same imported energy shock lifting everyone; the impulse is cost-push rather than a genuine pickup in demand. The honest gauge, the GDP deflator, has now been negative for a twelfth straight quarter, the longest stretch since China's market transition. Property is still contracting (development investment −16.2% year-to-date); retail sales limp along; the yuan is appreciating (~6.79 to the dollar) rather than weakening, which reads as a capital-outflow scare that has faded, not as reflation taking hold.

The transmission to the world runs through trade. Exports rose 19.4% in May to a record, and China is pushing surplus capacity (EVs at record export volumes, solar priced through the floor) into global markets. That is a disinflationary force pressing against the regime break everywhere else, and the reason the tariff wall keeps rising.

Add up the 2026 capital-spending plans of Amazon, Microsoft, Alphabet and Meta and you reach about $725bn, a 77% jump on 2025 and more than the annual GDP of most G20 members. Add Oracle and the number pushes past $780bn; the sell-side already models the "big five" above $1tn in 2027. It is the largest private capital-formation wave in corporate history, and it has run into a physical constraint.

That constraint is electricity. Chips can be fabricated and buildings can be poured; the hard part is firm power, delivered to a substation and co-located with compute, on the timescale of an AI arms race. The International Energy Agency expects global data-centre electricity demand to roughly double to 950 TWh by 2030. In the United States, Goldman sees data-centre load rising from ~31 GW to 66 GW by 2027; PJM, the largest US grid, attributes 94% of its 2024-2030 peak-load growth to data centres.

That has a second-order consequence that runs through this entire issue: the AI buildout is itself an inflationary force. US retail electricity rose 7% in 2025; the Dallas Fed estimates data centres add 0.05 percentage points to headline PCE this year, rising to 0.13pp by 2030. The compute boom is helping to keep central banks as hawkish as Part I described.

The defining corporate response is consolidation for scale. NextEra's ~$67bn all-stock bid for Dominion (announced 18 May, though the fight over it is very much a June story) would create the world's largest regulated electric utility: ~$249bn of market value, ~110 GW of capacity, and, critically, a monopoly over Northern Virginia's "Data Center Alley," the densest concentration of data centres on earth. In June the politics arrived: Senator Angus King urged FERC to reject it; Senate antitrust Democrats vowed scrutiny; Virginia legislators raised affordability alarms. The deal is not blocked, but it is now the test case for whether the state will let the grid consolidate around AI demand.

Below the merger, the contracts fill in the picture. Chevron signed a 20-year PPA with Microsoft for a co-located 2.67 GW gas plant in West Texas. Tesla, Sunrun and Renew Home aggregated 16+ GW of home batteries into a virtual power plant for hyperscale data centres. Google bought Intersect Power for $4.75bn to secure gigawatts directly. When the grid cannot deliver fast enough, buyers are securing their own supply.

The regulator moved too. On 18 June, FERC issued six "show cause" orders directing every US grid operator to justify or reform how they connect large loads, an admission that the interconnection regime cannot cope. PJM's 2027/28 capacity auction had already cleared near its cap at $333/MW-day and, for the first time, fell short of the entire region's reliability requirement.

Morgan Stanley estimates a ~$2.9tn global data-centre buildout through 2028, of which operating cash flow funds roughly half, leaving a ~$1.5tn financing gap. Its 2026 forecast: $250-300bn of debt from hyperscalers and their joint ventures, and ~$570bn of AI-related issuance all-in, more than double last year. By 31 May, $236bn had already priced, four times the year-earlier pace.

June showed the engine's newest gears. CoreWeave sold $3.55bn of dual-currency junk bonds, including the first-ever euro high-yield deal from a US AI-infrastructure company, with an order book above €7bn. Apollo and Blackstone closed a $35bn private-credit facility for Anthropic's compute: a special-purpose vehicle buys Google TPUs and leases them back, which keeps the debt off Anthropic's balance sheet, with Broadcom residual-value guarantees on $30bn of it. Data-centre asset-backed issuance is running toward JPMorgan's $30-40bn annual estimate.

The risk is concentration and structure. When compute is financed through SPVs, leases and residual-value guarantees, the leverage is real but the disclosure is thin. Oracle became the cautionary tale: after revealing ~$130bn of debt against its AI push, its stock had its worst week since 2001.

The data point the bulls had been waiting for landed on 25 June: global AI sales excluding China reached $25bn in Q1 2026, exceeding the ~$21bn of quarterly depreciation on the data centres and chips behind them, the second straight quarter above the line. Trailing-twelve-month generative-AI revenue hit $110bn, roughly three times faster than the internet, mobile or cloud waves grew at the same stage. The economics, for now, hold.

The caveat is in the word "barely." Depreciation still consumes more than two-thirds of AI revenue, the estimate excludes China, and it assumes a six-year life for hardware that may age faster. As the report's author, Azeem Azhar, put it: "the margin for error is narrow."

The market spent June testing that margin. On 5 June, a soft Broadcom outlook triggered a 10.3% one-day drop in the Philadelphia semiconductor index, its worst since March 2020, and erased over $1tn of chip value. Nvidia, which had touched $5.5tn in May, fell back toward $4.85tn. The Magnificent Seven lost roughly $2tn on the month and turned negative for the year (−3.3%), lagging the broad market for the first time since 2022. "Yet another way in which 2026 is looking like 1999," one strategist observed.

Why the layer matters more than the name

The AI trade the market chases is the model-builder and the chip. The trade that actually clears is the one that owns firm, dispatchable power co-located with compute. Demand is not the question: the IEA sees global data-centre electricity roughly doubling to 950 TWh by 2030, and PJM attributes 94% of its 2024-2030 peak-load growth to data centres. Supply is the question, and every layer above prices a different piece of that shortage.

The cleanest signals are in the plumbing, not the equities everyone quotes. Gas-turbine costs up 66% with five-year lead times is a backlog story for the OEMs. A regulator issuing six show-cause orders in a single day is an admission that interconnection, not generation, is the choke point, which is the transmission trade. And a 2.67 GW gas plant contracted straight to a hyperscaler is the co-location model that bypasses the grid entirely.

India runs the same playbook one layer down the cost curve. Its sovereign-compute build needs the same firm green power, and every gas-fired megawatt widens the oil-import bill the regime break is already punishing. That makes contracted renewables plus storage, wrapped in an InvIT at a 6.8 to 8.9% yield, the highest-leverage domestic expression of the whole thesis.

India imports about 88% of its crude, and roughly 40% of it transits the Strait of Hormuz. When the strait closed and Brent ran above $120, the pass-through was mechanical and brutal. May's petroleum-import bill hit a record ~$22.7bn, up 62% year-on-year; the combined April and May FY27 oil bill jumped 70%. The rupee fell to an all-time low of 96.9 on 20 May, and forex reserves have drawn down ~$56bn from their February peak of $728bn to ~$673bn.

June brought relief as Brent collapsed to the low $70s: the rupee recovered to ~94.6 and the 10-year government bond yield fell to 6.70%, a 15-week low. But the current-account math still points the wrong way: CRISIL projects the FY27 deficit widening to ~2.2% of GDP on costlier energy, from a FY26 deficit of just 0.6%. The one clean external-account print, a Q4 FY26 surplus of $7.1bn, predates the shock's pass-through, which will land in the July to September data.

India's response was to diversify the barrel. Russian crude hit a record ~2.70 mb/d in June (above half of all imports) as discounted, non-Hormuz supply was rushed in. New Delhi says it holds 76 to 80 days of fuel cover. India can reroute around the strait, but it cannot remove the exposure.

While the ECB, BoJ, Fed and BoE all leaned hawkish within a fortnight, India's Monetary Policy Committee held the repo rate at 5.25% on a unanimous 6-0 vote and kept its stance neutral. Governor Sanjay Malhotra was explicit: "If we wanted to prepare the market for rate hikes, we would have changed stance from neutral to restrictive." Hikes, he said, would be "premature."

Far from complacency, the hold was a calculated bet. The RBI raised its FY27 inflation forecast to 5.1% (from 4.6%) and cut growth to 6.6% (from 6.9%), acknowledging the shock. But it wagered that a falling oil price would do the tightening for it, and chose to defend the currency through the balance sheet rather than the policy rate. Alongside the hold came a ~$40bn package of inflow measures: a $5bn FX swap, an FCNR(B) hedging subsidy, expanded FAR access to long-dated G-secs, and the removal of the FPI general-route debt limit. Having sold a record $53bn defending the rupee in FY26, the RBI switched to buying dollars in June to rebuild reserves.

The early scoreboard favours the hold: oil fell, the rupee steadied, bonds rallied, GDP printed 7.7% for FY26, and foreign money flooded into Indian debt. The risk is symmetric. If the regime break proves durable and the dollar keeps climbing, a 5.25% rate against a 3.8% Fed dot and a rising DXY leaves the rupee exposed, and the RBI with a smaller reserve buffer than it started the year with.

India's external account is stronger than its headlines suggest and more fragile than its reserves imply. The FY26 current-account deficit was a benign 0.6% of GDP, but the overall balance of payments swung to a $23.6bn deficit, because the capital account thinned: net FDI was just $6.9bn (gross inflows a record ~$94.5bn, gutted by repatriation), and foreign portfolio investors pulled $16.4bn. What plugged the gap was borrowed and deposited money: $14.2bn of external commercial borrowing and $14.4bn of NRI deposits.

That is the vulnerability the oil shock exposed. Forecasts for the FY27 deficit span 1.1% (Goldman, on cheap oil) to 2.5% of GDP (Morgan Stanley, on $95 crude), and the BoP outcome splits the Street between a modest surplus and a $60bn+ deficit. Barclays' verdict is the one to hold: India needs ~$7bn to $8bn of inflows every month to balance, and the RBI's measures "will not resolve the structural BoP issues."

Hence the ~$40bn inflow package (FCNR deposit subsidies, FAR bond access, the removal of FPI debt limits, a $5bn swap). The catch: it all landed between 8 and 17 June, so none of it shows in the data yet, and the pre-package base was weak (FCNR(B) drew just $0.9bn in all of FY26). It is a bet that a cheaper oil bill and engineered inflows can buy time while the structural deficit remains unresolved.

The headline is a record exodus: foreign portfolio investors sold ₹2.74 lakh crore (~$29bn) of Indian equities in the first half of 2026, already more than in all of 2025, which dragged FPI ownership of NSE-listed companies to 15.8%, a 17-year low. For the first time, domestic institutions own more of the Nifty 500 than foreigners do.

The second line tells the real story. In June, even as FPIs sold ₹49,340 crore of equities, they bought ₹55,518 crore of Indian debt. That was the first month bond inflows exceeded equity outflows, and it took 2026 debt inflows to ₹63,784 crore. The RBI's FAR expansion and the removal of the FPI debt limit, plus a bond yield falling to 6.70%, made Indian duration the trade. The same foreign capital that fled the equity tape arrived in the government-bond market.

Underneath both sits the domestic floor. Domestic institutions bought ~₹4.7 lakh crore of equities in H1, which absorbed ~90% of FPI selling in FY26. The buying was funded by SIP inflows running at a record ~₹31,000 crore a month and a mutual-fund industry now at ₹81.6 lakh crore. The indices bent (Nifty −8.4%, Sensex −10.3% in H1) but did not break.

The Sensex closed June at 76,479 and the Nifty at 23,866, down 10.3% and 8.4% respectively for the half, the worst first half for the Sensex since the pandemic, and roughly 9 to 11% below the September-2024 record. Yet both rose modestly in June as oil fell and bonds rallied, and India's market capitalisation crept back above $5tn.

The damage was concentrated. Nifty IT fell about 31% in H1 (its worst first half since 2003) as fear that generative AI will compress the services model gutted the majors: Infosys −38%, TCS −37%, Wipro −35%. This is the domestic face of the global AI trade, and we give it its own page overleaf. Money rotated hard into defensives and domestic-facing names (realty, FMCG, pharma, chemicals and defence all outperformed) while breadth stayed strong beneath the weak headline: the Nifty Microcap 250 rose 11.4% in H1, the best of any tier.

Valuations sit near, not above, the long-run average (the Nifty's trailing P/E around 20.8×), and India's premium to emerging markets has compressed from its 2024 extreme. After a record foreign exodus, the domestic bid has kept the market expensive-ish rather than cheap. That is the tension a value buyer must resolve.

India lives on both sides of the AI trade, and in June the two sides pulled apart. Abroad, AI is a capital-formation supercycle: the hyperscaler capex, the power scramble, the debt engine of Part III. At home, the first-order effect on India's flagship export sector has been destruction: the IT-services majors, whose model is selling human engineering hours, fell about a third in the first half as the market concluded that agentic AI erodes the very billable hour they sell.

That is the paradox. The country that supplies the world's software talent is, for now, a net loser in the equity market from software's biggest leap, while the country that owns the chips, the clouds and the power is the winner. India's answer, the sovereign-compute build this magazine chronicled in Issue 01, now runs into the same constraint the Americans hit: power. India's own data-centre ambition (Reliance, Adani, the hyperscalers' local regions) competes for grid capacity, and its renewable-plus-storage thesis is the only credible way to feed it without widening the very oil-and-gas import bill that Hormuz just inflated.

So the trades invert. Short the narrative that Indian IT services are cheap because they've fallen; the de-rating may be structural, not cyclical. Long the picks-and-shovels of India's own compute build (power, transmission, data-centre real estate, and the semiconductor packaging layer), where AI is a tailwind rather than a threat.

Higher-for-longer did not freeze the deal market. It changed its shape. Bain's mid-year review put global deal value up 41% year-on-year in the first five months, on pace for roughly $5.3tn, a whisker below the 2020 record. PwC sees ~$4tn, +13%, the strongest since 2021. The common thread in every provider's data is concentration: megadeals (>$5-10bn) are taking a record share, while total deal counts fall. When money costs more, acquirers make fewer, more strategic bets, and pay in stock more often (a record ~35% cash-and-stock mix).

Private equity is sitting on a wall of dry powder and finally deploying it into scaled platforms. In India, a queue of billion-dollar processes formed in June: KKR, Warburg, TPG, Advent, CVC and Permira circled the maternity chain Cloudnine at a ~$1bn valuation; Blackstone, Advent and Warburg are chasing Axis Finance at a similar size. The exit drought that constrained Indian PE is easing as strategic buyers and a reopening IPO window provide the way out.

The primary pipeline is the release valve. India has 236 mainboard DRHPs in the queue and 23 issuers have already raised over ₹27,000cr in 2026; a ~$48bn wave of IPO lock-in expiries across 79 stocks is redistributing ownership from founders and early backers to the public float. The domestic-institution bid, record SIPs, is the natural absorber.

Higher-for-longer is a tax on the leveraged-buyout model. A decade ago a deal needed ~5% annual EBITDA growth to hit a 2.5x return; with entry multiples at a record 14x, leverage cut from 50% to 36% of enterprise value, and debt costing 8-9%, Bain calculates the same return now needs 10-12%. "Twelve is the new five." With financial engineering no longer enough, the return now has to be earned through operations.

The bigger problem is the exit. Distributions have run below 15% of net asset value for four straight years, a record. That has left ~32,000 unsold companies worth ~$3.8tn, and average holding periods have stretched to seven years. With the IPO window still barely open for sponsors (just five PE-backed listings in Q1), the industry is manufacturing liquidity synthetically: a record ~$230bn secondaries market, continuation vehicles now ~43% of it, and ~$150bn of NAV loans. LPs have noticed: DPI is now the top re-up criterion, and roughly one in five is cutting buyout allocations.

Private credit, the decade's great absorber, showed its first cracks. AUM cleared $2tn, but Fitch's "true" default rate hit a record 5.8%, PIK interest doubled to ~11% of the market, and in June Blackstone's BCRED joined funds capping redemptions. The plumbing that financed the buyout and AI booms is being stress-tested for the first time.

Futures put a hike by October at ~60%, and Goldman has removed every 2026 cut. The regime's signature is in the bond math: 10-year real yields rose to ~2.2% while breakevens stayed anchored near 2.2%. The market is repricing real rates, not inflation expectations. That is the higher-and-energy-scarce world, priced into the bond market.

History is the swing factor. Central banks that hiked into a supply shock have often reversed within one-to-two quarters once it faded, as the ECB did in 2008 and 2011. That is the bull case. But the 1970s warn that easing before inflation is anchored re-entrenches it. That is the bear case. The base case is that the Fed simply waits, data-dependent, through the 29 July meeting.

The through-line is simple: own what a higher, flatter, energy-scarce rate path rewards. That means real assets, power, positive real yield, and the one central bank that broke ranks (India, via its bond market). Fund it by underweighting what the old reflex priced: long-duration growth, the disinflation hedges bid for a rate-cut world, and the AI credit whose returns assume everything goes right into a regime that just turned against it.

Private-markets overlay: follow the smart money into real assets and private credit, but demand collateral and daily liquidity; the June redemption caps are the warning. Underwrite the AI buildout through the power layer (12-18% target IRRs on contracted infra) rather than the model-builders or their junk-rated debt.

Q. Is one hawkish week really a "regime"? Central banks reverse constantly.

They do, which is why we anchor the claim on behaviour, not rhetoric. The evidence is that four banks tightened and then held as the oil trigger collapsed. A reversible pivot reverses when the shock fades; this one didn't. That is the signal.

Q. If equities keep making records, why reposition at all?

Because the leadership is changing under the surface. The Mag 7 turned negative on the year for the first time since 2022 while power, defence, European and Japanese equities led. You don't fight the index; you rotate within it, toward the assets the new rate path rewards.

Q. The RBI held. Is Indian duration a buy or a trap?

A buy with a currency hedge. FAR-eligible G-secs at ~6.7% with a central bank on hold and a falling oil bill are attractive. The foreign bid proved it in June. The trap is unhedged: if the dollar breaks out, the rupee gives back what the coupon pays.

The Regime Break · Capital Axis Framework

The cover thesis of this issue: the June 2026 shift in which the world's anchor central banks retired the disinflation-era reaction function. They moved hawkish in concert and refused to ease when the triggering oil shock collapsed. The change is in behaviour, not the rate level.

Higher-and-Energy-Scarce

Our label for the successor regime: rates settle high and energy is a recurring source of inflation shocks, so central banks act earlier and ease only on realised (not forecast) disinflation. The evolution of "higher-for-longer" from a level to a rule.

The Reaction-Function Reset

The specific mechanism of the break is that the market re-prices the response rule rather than the forecast, that is, what the central bank will do the next time something breaks. It is measured by the correlation of policy moves rather than their magnitude.

The Power Bill

The thesis that the binding constraint on the AI supercycle is firm electricity rather than silicon, and that the buildout is itself an inflationary force that feeds back into the regime break. Own the watts, not the models.